services/nodemanager/arvnodeman/nodelist.py

   1 #!/usr/bin/env python
   2 # Copyright (C) The Arvados Authors. All rights reserved.
   3 #
   4 # SPDX-License-Identifier: AGPL-3.0
   5
   6 from __future__ import absolute_import, print_function
   7
   8 import subprocess
   9
  10 from . import clientactor
  11 from . import config
  12
  13 import arvados.util
  14
  15 class ArvadosNodeListMonitorActor(clientactor.RemotePollLoopActor):
  16     """Actor to poll the Arvados node list.
  17
  18     This actor regularly polls the list of Arvados node records, and
  19     sends it to subscribers.
  20     """
  21
  22     def is_common_error(self, exception):
  23         return isinstance(exception, config.ARVADOS_ERRORS)
  24
  25     def _item_key(self, node):
  26         return node['uuid']
  27
  28     def _send_request(self):
  29         nodelist = arvados.util.list_all(self._client.nodes().list)
  30
  31         # node hostname, state
  32         sinfo_out = subprocess.check_output(["sinfo", "--noheader", "--format=%n %t"])
  33         nodestates = {}
  34         for out in sinfo_out.splitlines():
  35             try:
  36                 nodename, state = out.split(" ", 2)
  37                 if state in ('alloc', 'alloc*',
  38                              'comp',  'comp*',
  39                              'mix',   'mix*',
  40                              'drng',  'drng*'):
  41                     nodestates[nodename] = 'busy'
  42                 elif state == 'idle':
  43                     nodestates[nodename] = 'idle'
  44                 else:
  45                     nodestates[nodename] = 'down'
  46             except ValueError:
  47                 pass
  48
  49         for n in nodelist:
  50             if n["slot_number"] and n["hostname"] and n["hostname"] in nodestates:
  51                 n["crunch_worker_state"] = nodestates[n["hostname"]]
  52             else:
  53                 n["crunch_worker_state"] = 'down'
  54
  55         return nodelist
  56
  57 class CloudNodeListMonitorActor(clientactor.RemotePollLoopActor):
  58     """Actor to poll the cloud node list.
  59
  60     This actor regularly polls the cloud to get a list of running compute
  61     nodes, and sends it to subscribers.
  62     """
  63
  64     def __init__(self, client, timer_actor, server_calc, *args, **kwargs):
  65         super(CloudNodeListMonitorActor, self).__init__(
  66             client, timer_actor, *args, **kwargs)
  67         self._calculator = server_calc
  68
  69     def is_common_error(self, exception):
  70         return isinstance(exception, config.CLOUD_ERRORS)
  71
  72     def _item_key(self, node):
  73         return node.id
  74
  75     def _send_request(self):
  76         nodes = self._client.list_nodes()
  77         for n in nodes:
  78             # Replace with libcloud NodeSize object with compatible
  79             # CloudSizeWrapper object which merges the size info reported from
  80             # the cloud with size information from the configuration file.
  81             n.size = self._calculator.find_size(n.size.id)
  82         return nodes